Going serverless with Rust and Shuttle

Mar 30, 2023

rust

530 views

In my epic quest to find any excuse to write more Rust in my daily dev life, I stumbled across an incredible platform that allows developers to write serverless functions entirely in Rust.

Serverless functions? Check.

Written in pure Rust? Check.

All managed from the comfort of the command line? Dear god... just take my money!

What the deployment story currently looks like

Harkening back to looking for any excuse to write more Rust, my typical deployment workflow for small projects would look something like this:

Write code (entirely bug-free, obviously)
Construct a sensible Dockefile
Deploy to some sort managed SaaS infrastructure platform
Configure CI, testing pipelines, etc.

While the current landscape of tools is small enough for micro-projects, I'd love to be an even lazier developer than I currently am and remove the middle to bullet points above. With shuttle, I'm able to write code and simply deploy my function - no containerization, no server configurations, simply just running a few commands and I'm able to go from local dev to production in minutes. Let's write a serverless function with shuttle that retrieves GitHub stars from one of our repositories!

Getting started

First, let's bootstrap our shuttle project using cargo-shuttle:

cargo install cargo-shuttle # or quickinstall if you prefer

Quick the shuttle CLI in place, let's scaffold out an axum server to respond to our requests:

cargo shuttle init --axum

Follow the prompts of naming the project and selecting a folder. Taking a look at main.rs that shuttle generates for us should look something like this:

use axum::{routing::get, Router};

async fn hello_world() -> &'static str {
    "Hello, world!"
}

#[shuttle_runtime::main]
async fn axum() -> shuttle_axum::ShuttleAxum {
    let router = Router::new().route("/hello", get(hello_world));

    Ok(router.into())
}

and our manifest file will look something along the lines of:

[package]
name = "serverless-rust-with-shuttle"
version = "0.1.0"
edition = "2021"
publish = false

# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html

[dependencies]
shuttle-runtime = "0.12.0"
axum = "0.6.12"
shuttle-axum = "0.12.0"
tokio = "1.26.0"

A few key notes here:

Shuttle bootstraps a few dependencies for us in shuttle-runtime and shuttle-axum (0.12.0 at the time of this writing)
These crates allow us to invoke our function as an axum-specific serverless application as we see in our main.rs file

But what's this #[shuttle_runtime::main] macro on our main function? Let's take a look with a quick cargo expand if you have it installed:

#![feature(prelude_import)]
#[prelude_import]
use std::prelude::rust_2021::*;
#[macro_use]
extern crate std;
use axum::{routing::get, Router};

async fn hello_world() -> &'static str { "Hello, world!" }

fn main() {
    let body = async { shuttle_runtime::start(loader).await; };

    #[allow(clippy :: expect_used, clippy :: diverging_sub_expression)]
    {
        return tokio::runtime::Builder::new_multi_thread().enable_all().build().expect("Failed building the Runtime").block_on(body);
    }
}
async fn loader(mut _factory: shuttle_runtime::ProvisionerFactory,
    logger: shuttle_runtime::Logger) -> shuttle_axum::ShuttleAxum {
    use shuttle_runtime::Context;
    use shuttle_runtime::tracing_subscriber::prelude::*;
    let filter_layer =
        shuttle_runtime::tracing_subscriber::EnvFilter::try_from_default_env().or_else(|_|
                    shuttle_runtime::tracing_subscriber::EnvFilter::try_new("INFO")).unwrap();
    shuttle_runtime::tracing_subscriber::registry().with(filter_layer).with(logger).init();
    axum().await
}
async fn axum() -> shuttle_axum::ShuttleAxum {
    let router = Router::new().route("/hello", get(hello_world));

    Ok(router.into())
}

Whoa! If we look closely, we see a familiar bit of generated code:

#[allow(clippy :: expect_used, clippy :: diverging_sub_expression)]
{
    return tokio::runtime::Builder::new_multi_thread().enable_all().build().expect("Failed building the Runtime").block_on(body);
}

Shuttle's #[shuttle_runtime::main] macro generates the same code a typical #[tokio::main] macro generates along with a few extra bits (like including the defacto logging crate in tracing for logging to the server's console through tracing-subscriber). No magic here, just a bit of generated boilerplate for us to hit the ground running.

A quick note, shuttle recently (at the time of this writing) added a dependency on protobuf to facilitate some of their internal infrastructure. If you're missing protoc as a dependency on your machine, take a look at my blog's README for some quick instructions on how to get up and running with the package.

Managing secrets

So we've set out to build a serverless function that will retrieve stars from various repositories in GitHub, which implies we'll need to interact with the GitHub API. To use their API, we'll need a key to authenticate our requests but shouldn't rely on clients, nor expose our secret keys to clients. We'll need some way to inject secrets at startup/runtime that'll house our API keys only known to the function so we can make valid requests - luckily shuttle has us covered with shuttle secrets.

Luckily, it's rather straightforward to secrets into our serverless function by adding a Secrets.toml file, shuttle's version of a .env file, and injecting them on startup with the shuttle-secrets crate. Let's add that to our cargo dependencies:

cargo add shuttle-secrets

and in our main function, let's add the secrets store as an argument (don't forget to add a use shuttle_secrets::SecretStore; to your imports):

#[shuttle_runtime::main]
async fn axum(
    #[shuttle_secrets::Secrets] secret_store: SecretStore,
) -> shuttle_axum::ShuttleAxum {
  // other code...
}

While we're at it, we'll need a valid GitHub access token - for creating personal access tokens, check out the docs and make sure to add the repo:read scope so our authenticated requests have permission to retrieve repository data.

Once you've got your token (save it off as you'll only be able to see it once on the UI), let's add a Secrets.toml file at the root of our project with the token as a key:

GITHUB_ACCESS_TOKEN = "ghp..."

Now that we've got secrets in place, let's do a sanity check to make sure we have access to our token value once the function is running. Let's update our main function to spit out the token's value at startup:

#[shuttle_runtime::main]
async fn axum(#[shuttle_secrets::Secrets] secret_store: SecretStore) -> shuttle_axum::ShuttleAxum {
    let token = secret_store
        .get("GITHUB_ACCESS_TOKEN")
        .expect("No access token was provided.");

    dbg!(token);

    let router = Router::new().route("/hello", get(hello_world));

    Ok(router.into())
}

We'll simply just .expect() the token to exist as it's required for us to operate. Using shuttle's CLI, let's spin this function up locally:

cargo shuttle run

and after our project is compiled and run, we should see the token's value in our terminal:

Finished dev [unoptimized + debuginfo] target(s) in 30.18s
[samples/serverless-rust-with-shuttle/src/main.rs:14] token = "ghp..."

Starting serverless-rust-with-shuttle on http://127.0.0.1:8000

Sweet! We're 90% ready to start writing the actual fun code. Since we'll need to call an external API via HTTP, let's add reqwest as a dependency to make our lives easier (with the json feature):

cargo add reqwest --features json

Okay... now we're ready to get the ball rolling. Let's update the existing route with a path parameter in the form of the repository name along with scaffolding out a handler in our main function:

use axum::{routing::get, Router};
use shuttle_secrets::SecretStore;

async fn get_repository_stars() -> &'static str {
    "Hello, world!"
}

#[shuttle_runtime::main]
async fn axum(#[shuttle_secrets::Secrets] secret_store: SecretStore) -> shuttle_axum::ShuttleAxum {
    let token = secret_store
        .get("GITHUB_ACCESS_TOKEN")
        .expect("No access token was provided.");

    let router = Router::new().route("/:repository/stars", get(get_repository_stars));

    Ok(router.into())
}

and with our server running, let's send through a request:

curl --location 'localhost:8000/my-repository/stars'
Hello, world!

Nice, now we're getting somewhere. Let's add cargo-watch to have our server restart anytime we make changes so we're not bothered to stop/start manually ourselves:

cargo install cargo-watch # or cargo binstall

and running our server again:

cargo watch -x 'shuttle run'

Now any change to our source code will trigger an automatic restart of our server. Let's update our function handler to return some JSON in a sane fashion so we can start stubbing out what the request flow will look like. Back in main.rs let's update our handler as well as add a response JSON model:

#[derive(Serialize, Debug)]
struct StarsResponse {
    count: usize,
}

async fn get_repository_stars() -> Result<Json<StarsResponse>, &'static str> {
    let response = StarsResponse { count: 9000 };
    Ok(Json(response))
}

Ignoring our errors for just a bit, we'll need to bring in serde so we can serialize our responses to JSON with the derive flag so we can use it on our struct:

cargo add serde --features derive

Compiling and running our server again, let's send another request through:

curl --location 'localhost:8000/my-repository/stars'
{"count":9000}

Now we're getting somewhere. We don't want to just return errors in the form &'static strs, so let's address that by defining our own error implementation that plays nicely with axum:

#[derive(Serialize, Debug)]
struct HandlerError {
    message: String,
}

enum ApiError {}

impl IntoResponse for ApiError {
    fn into_response(self) -> axum::response::Response {
        let (status, error_message) = match self {
            _ => (
                StatusCode::INTERNAL_SERVER_ERROR,
                "Something very, very bad has happened... :(".to_string(),
            ),
        };

        let body = Json(HandlerError {
            message: error_message,
        });

        (status, body).into_response()
    }
}

async fn get_repository_stars() -> Result<Json<StarsResponse>, ApiError> {
    let response = StarsResponse { count: 9000 };
    Ok(Json(response))
}

Ignoring clippy errors for a moment, let's quickly run through our new additions:

We've added a HandlerError struct to give us the ability to transform errors into user-friendly JSON responses with a message
We've impl'd InotoResponse for our new ApiError type that we'll use to coerce errors that happen during the request into something axum understands how to transform
We've updated our handler to return an ApiError in our Result rather than a static string so we can again help our users out with information about why the request failed

We'll get around to adding some branches to our ApiError enum eventually, but for now to get our code to compile, let's add the http crate so we can lean on the StatusCode type to map internal handler errors to sensible HTTP status codes.

Handling errors with axum deserves it's on blog post, so I'll gloss over a few of the details for now so we can focus on just getting our function up and running.

cargo add http

Now running our code and making another request, we should still see the same message response as the previous request we made before we updated our handler. Our main.rs file is getting rather large, so let's split some things out for organizational purposes. Let's add two additional files in errors.rs and handlers.rs to house our error implementation and request handlers, respectively

errors.rs

use axum::{response::IntoResponse, Json};
use http::StatusCode;
use serde::Serialize;

#[derive(Serialize, Debug)]
struct HandlerError {
    message: String,
}

pub enum ApiError {}

impl IntoResponse for ApiError {
    fn into_response(self) -> axum::response::Response {
        let (status, error_message) = match self {
            _ => (
                StatusCode::INTERNAL_SERVER_ERROR,
                "Something very, very bad has happened... :(".to_string(),
            ),
        };

        let body = Json(HandlerError {
            message: error_message,
        });

        (status, body).into_response()
    }
}

handlers.rs

use axum::Json;
use serde::Serialize;

use crate::errors::ApiError;

#[derive(Serialize, Debug)]
pub struct StarsResponse {
    count: usize,
}

pub async fn get_repository_stars() -> Result<Json<StarsResponse>, ApiError> {
    let response = StarsResponse { count: 9000 };
    Ok(Json(response))
}

After cleaning up a few import errors and peppering in a few pubs for visibility, our main.rs file should now look like this:

mod errors;
mod handlers;

use axum::{routing::get, Router};
use handlers::get_repository_stars;
use shuttle_secrets::SecretStore;

#[shuttle_runtime::main]
async fn axum(#[shuttle_secrets::Secrets] secret_store: SecretStore) -> shuttle_axum::ShuttleAxum {
    let token = secret_store
        .get("GITHUB_ACCESS_TOKEN")
        .expect("No access token was provided.");

    let router = Router::new().route("/:repository/stars", get(get_repository_stars));

    Ok(router.into())
}

Nice and lean! Next, we need to somehow get our token into our get_repository_stars handler in a sane fashion, and it would be nice to inject it as a dependency as it represents part of our serverless function's state. Let's lean on axum's state management mechanisms to do so with the .with_state() router extensions so we can inject the token secret into the handler at request time.

There are lots of ways to facilitate handling state in axum, so let's take the simplest approach by wrapping our token in a struct that'll we'll inject into our handlers. To keep things clean, let's add a state.rs file and the necessary bits to get a bit of thread-safe state working:

use std::sync::Arc;

#[derive(Debug)]
pub struct HandlerState {
    pub access_token: String,
}

impl HandlerState {
    pub fn new_state(access_token: String) -> Arc<HandlerState> {
        Arc::new(HandlerState { access_token })
    }
}

and back in our router in main.rs, let's add it as a state extension:

#[shuttle_runtime::main]
async fn axum(#[shuttle_secrets::Secrets] secret_store: SecretStore) -> shuttle_axum::ShuttleAxum {
    let token = secret_store
        .get("GITHUB_ACCESS_TOKEN")
        .expect("No access token was provided.");

    let router = Router::new()
        .route("/:repository/stars", get(get_repository_stars))
        .with_state(HandlerState::new_state(token));

    Ok(router.into())
}

Let's do one more sanity check to make sure we're still able to access our token within our request handler. Back in our get_repository_stars handler:

pub async fn get_repository_stars(
    State(state): State<Arc<HandlerState>>,
) -> Result<Json<StarsResponse>, ApiError> {
    dbg!(&state.access_token);
    let response = StarsResponse { count: 9000 };
    Ok(Json(response))
}

and thanks to cargo-watch, our server should be back up and running. Sending through yet another request, we should see our server output something like:

&state.access_token = "ghp..."

Nice! We're propagating down our state to our handler leaning on Arc to help us facilitate sharing our HandlerState across request threads and we're finally in a spot to start calling out to the GitHub API.

Before we do so, let's take a look at what we've got so far:

We've got a serverless function spun with axum bootstrapped with shuttle
We're handling errors according to axum convention
We've separated out our bits of code into logically grouped units
We're propagating top-level application state safely down to request handlers

Doesn't seem like much, but we've accomplished quite a bit! Let's go back and add a bit of tracing so we can see inside the mind of our function as it processes requests. Recall earlier in the expanded macro just above main that shuttle provides we have our application bootstrapped with tracing behind the scenes ready to go to start logging. Let's add some trace logging in a few places so we can pretty-print out to the console. First, let's add the tracing crate:

cargo add tracing

Next, let's update our main function to include a bit of startup-style logging:

#[shuttle_runtime::main]
async fn axum(#[shuttle_secrets::Secrets] secret_store: SecretStore) -> shuttle_axum::ShuttleAxum {
    tracing::info!("Bootstrapping function secrets");

    let token = secret_store
        .get("GITHUB_ACCESS_TOKEN")
        .expect("No access token was provided.");

    tracing::info!("Secrets successfully read, building server router");

    let router = Router::new()
        .route("/:repository/stars", get(get_repository_stars))
        .with_state(HandlerState::new_state(token));

    tracing::info!("Router successfully initialized, now listening on port 8000");

    Ok(router.into())
}

Then, let's add some logging to our request handler in handlers.rs:

pub async fn get_repository_stars(
    State(state): State<Arc<HandlerState>>,
    Path(repository): Path<String>,
) -> Result<Json<StarsResponse>, ApiError> {
    tracing::info!(
        "Received request to get start count for repository {}",
        repository
    );
    let response = StarsResponse { count: 9000 };
    Ok(Json(response))
}

Notice that I've added a Path(repository): Path<String> to our handler's input - this is axum's way of accessing dynamic route values based on the :repository path route from our router. Spinning up our function (or simply waiting if you're still cargo watching), we should the output in the console (omitting timestamps for brevity):

INFO serverless_rust_with_shuttle: Bootstrapping function secrets
INFO serverless_rust_with_shuttle: Secrets successfully read, building server router
INFO serverless_rust_with_shuttle: Router successfully initialized, now listening on port 8000

Starting serverless-rust-with-shuttle on http://127.0.0.1:8000

Sending a request through again, we should see the repository name from the route path being logged as well:

# From another terminal window...
curl --location 'localhost:8000/my-repository/stars'

# In our output console
INFO serverless_rust_with_shuttle::handlers: Received request to get start count for repository my-repository

Nice! Pat yourself on the back, we've written quite a bit of code and have a functioning server listening to requests. We still have two things left:

Calling the GitHub API to retrieve star counts
Deploying our function to production

Deploying to shuttle

Before we finish up the core logic of our code, let's left-shift our deployment process a bit. Let's deploy our function, verifying all our I's are dotted and T's crossed. Heading over to shuttle's website and signing up, we should be met with a screen like the following:

shuttle dashboard

I'm also going to update the name of our function to something a bit more relevant with a Shuttle.toml file at the root of our project:

name = "github-repository-star-counter"

You'll need to adjust the name as, sadly, I'll be taking this name for myself. Once you're authenticated, simply run the deploy command cargo shuttle deploy and we should a bunch of internal logging from shuttle along with a successful deploy message along the lines of:

These secrets can be accessed by github-repository-star-counter
╭─────────────────────╮
│         Keys        │
╞═════════════════════╡
│ GITHUB_ACCESS_TOKEN │
╰─────────────────────╯

Service Name:  github-repository-star-counter
Deployment ID: 3339ef4c-60f0-47e6-a159-5034ac03ad4f
Status:        running
Last Updated:  2023-03-29T23:09:40Z
URI:           https://github-repository-star-counter.shuttleapp.rs

Heck yeah! Our function has been deployed and also picked up our key from our Secrets.toml file. Let's test it out by curling to the URI:

curl --location https://github-repository-star-counter.shuttleapp.rs/my-repository/stars
{"count":9000}

We've officially got serverless Rust running in production - how cool is that?

With our initial deployment out of the way, let's finish fleshing out our function to retrieve repository stars.

Back to business

Let's add the client request to GitHub. Since we'll be establishing a connection to GitHub's API servers, rather than spin up a new HTTP client per request, let's instantiate a single client at startup for our handlers to pull out of from state. There are lots of benefits to recycling HTTP client connections throughout an application's lifetime, but that's a bit beyond the scope of what we're doing today.

Let's update our HandlerState to include a Client from the reqwest crate:

use std::sync::Arc;

use reqwest::Client;

#[derive(Debug)]
pub struct HandlerState {
    pub access_token: String,
    pub client: Client,
}

impl HandlerState {
    pub fn new_state(access_token: String) -> Arc<HandlerState> {
        let client = Client::new();

        Arc::new(HandlerState {
            access_token,
            client,
        })
    }
}

Now that we'll have access to the HTTP client, let's test out a call to the repositories. The URL we'll be calling to retrieve repository information will be in the form of https://api.github.com/repos/OWNER/REPO where we'll hard code OWNER to be your username for now. Let's test a call out to see what the response looks like:

curl --request GET \
--url "https://api.github.com/repos/joeymckenzie/realworld-rust-axum-sqlx" \
--header "Accept: application/vnd.github+json" \
--header "Authorization: Bearer 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"
{
  "id": 485222387,
  "node_id": "R_kgDOHOvn8w",
  "name": "realworld-rust-axum-sqlx",
  // ...a ton of other properties
  "stargazers_count": 129,
}

We see in the response we get a lot of other data that doesn't necessarily pertain to the number of stars on the repositories. The only property we care about for now is the stargazers_count which represents the number of stars our repository has. Let's create a response model to deserialize this response into Rust code. Back in our handlers.rs file:

// Imports...

#[derive(Serialize, Debug)]
pub struct StarsResponse {
    count: usize,
}

#[derive(Deserialize, Debug)]
pub struct GitHubRepositoryResponse {
    stargazers_count: usize
}

pub async fn get_repository_stars(
    State(state): State<Arc<HandlerState>>,
    Path(repository): Path<String>,
) -> Result<Json<StarsResponse>, ApiError> {
    // Code...
}

We've added a GithubRepositoryResponse to handle pulling data out into a struct. Let's implement the call now in our handler:

pub async fn get_repository_stars(
    State(state): State<Arc<HandlerState>>,
    Path(repository): Path<String>,
) -> Result<Json<StarsResponse>, ApiError> {
    tracing::info!(
        "Received request to get start count for repository {}",
        repository
    );

    let url = format!("https://api.github.com/repos/joeymckenzie/{}", repository);

    let response = state
        .client
        .get(url)
        .bearer_auth(state.access_token)
        .send()
        .await?;

    let response = StarsResponse { count: 9000 };

    Ok(Json(response))
}

Now if try to compile, we'll get an error yelling at us stating we have no conversion between a reqwest error and something axum understands with our ApiError. Yep, you guessed it - time to do some error converting.

Propagating errors

One of the first fundamental concepts we learn in Rust is the proper handling and conversion of errors. With Rust's expansive crate ecosystem and library authors offering custom errors about their internal processes in their public APIs, we're bound to eventually be forced to convert external errors to something that's known within our programs. There are many great articles, including the book itself, that do a much more fantastic job than I'll do here going over converting errors from (in our case) reqwest to our internal ApiError.

In the simplest case, we'll need to convert From a reqwest::Error to an ApiError:

errors.rs

pub enum ApiError {
    RequestFailed(String),
}

impl From<reqwest::Error> for ApiError {
    fn from(value: reqwest::Error) -> Self {
        Self::RequestFailed(value.to_string())
    }
}

We could do this for any number of errors that will rear their ugly heads at some point while our function is executing, but for now, I'll take the easy way out with thiserror:

cargo add thiserror

Now we can trim up errors.rs:

use axum::{response::IntoResponse, Json};
use http::StatusCode;
use serde::Serialize;
use thiserror::Error;

#[derive(Serialize, Debug)]
struct HandlerError {
    message: String,
}

#[derive(Error, Debug)]
pub enum ApiError {
    #[error("The request to GitHub failed: {0}")]
    RequestFailed(#[from] reqwest::Error),
}

impl IntoResponse for ApiError {
    fn into_response(self) -> axum::response::Response {
        let (status, error_message) = match self {
            Self::RequestFailed(e) => (StatusCode::INTERNAL_SERVER_ERROR, e.to_string()),
        };

        let body = Json(HandlerError {
            message: error_message,
        });

        (status, body).into_response()
    }
}

Leaning on thiserror, we can leverage the #[error] macro to spit out a bit of boilerplate error conversion code for us. For a sanity check, let's take a look at the generated code with another quick cargo expand errors:

pub enum ApiError {
    #[error("The request to GitHub failed")]
    RequestFailed(#[from] reqwest::Error),
}

#[allow(unused_qualifications)]
impl std::error::Error for ApiError {
    fn source(&self) -> std::option::Option<&(dyn std::error::Error + 'static)> {
        use thiserror::__private::AsDynError;
        #[allow(deprecated)]
        match self {
            ApiError::RequestFailed { 0: source, .. } => {
                std::option::Option::Some(source.as_dyn_error())
            }
        }
    }
}

#[allow(unused_qualifications)]
impl std::fmt::Display for ApiError {
    fn fmt(&self, __formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
        #[allow(unused_variables, deprecated, clippy::used_underscore_binding)]
        match self {
            ApiError::RequestFailed(_0) => {
                __formatter.write_fmt(format_args!("The request to GitHub failed"))
            }
        }
    }
}

#[allow(unused_qualifications)]
impl std::convert::From<reqwest::Error> for ApiError {
    #[allow(deprecated)]
    fn from(source: reqwest::Error) -> Self {
        ApiError::RequestFailed {
            0: source,
        }
    }
}

#[automatically_derived]
impl ::core::fmt::Debug for ApiError {
    fn fmt(&self, f: &mut ::core::fmt::Formatter) -> ::core::fmt::Result {
        match self {
            ApiError::RequestFailed(__self_0) => {
                ::core::fmt::Formatter::debug_tuple_field1_finish(
                    f,
                    "RequestFailed",
                    &__self_0,
                )
            }
        }
    }
}

Sifting through the other bits of code that are printed out to the console, we see that thiserror is generating some boilerplate to impl std::error::Error and From<reqwest::Error> for us, so we can avoid writing the implementations ourselves. Thanks, dtolnay!

So many stars

Okay, back to our handler. So we're handling the result errors reqwest could propagate back to us, now let's rip the stargazer_count off the API response to map back to our StarsResponse struct. Let's throw a .json() after our .await? to do so:

pub async fn get_repository_stars(
    State(state): State<Arc<HandlerState>>,
    Path(repository): Path<String>,
) -> Result<Json<StarsResponse>, ApiError> {
    tracing::info!(
        "Received request to get start count for repository {}",
        repository
    );

    let url = format!("https://api.github.com/repos/joeymckenzie/{}", repository);
    dbg!(url.clone());

    let github_response = state
        .client
        .get(url)
        .bearer_auth(&state.access_token)
        .header("User-Agent", "github-repository-star-counter/0.0.1")
        .send()
        .await?
        .json::<GitHubRepositoryResponse>()
        .await?;

    let response = StarsResponse {
        count: github_response.stargazers_count,
    };

    Ok(Json(response))
}

We'll attempt to deserialize the response into our GitHubRepositoryResponse and again await? the process as we need to read from the response buffer and propagate any errors. We're already converting between reqwest errors and our internal ApiError, so we're all good there.

We also add a User-Agent header to let the GitHub API servers know who we are - this is arbitrary for our purposes, but is important for requests coming from the browser, in Postman, etc. Let's spin up our function and send a request through:

 curl -l http://localhost:8000/realworld-rust-axum-sqlx/stars
{"count":129}

We have a response! Now that we've got the core logic in place, let's go ahead and deploy our function with cargo shuttle deploy. Once the deployment finishes, let's ping our function at the deployment URL:

curl -l https://github-repository-star-counter.shuttleapp.rs/realworld-rust-axum-sqlx/stars
{"count":129}

Nice! We've got good responses coming back from a serverless function written entirely in Rust. Let's check the logs with a quick cargo shuttle logs to trace our request:

cargo shuttle logs
# A few other logs that aren't important for now...
 INFO serverless_rust_with_shuttle::handlers: Received request to get start count for repository realworld-rust-axum-sqlx
DEBUG reqwest::connect: starting new connection: https://api.github.com/
DEBUG hyper::client::connect::dns: resolving host="api.github.com"
DEBUG hyper::client::connect::http: connecting to 140.82.121.6:443
 INFO serverless_rust_with_shuttle::handlers: Response received from GitHub GitHubRepositoryResponse { stargazers_count: 129 }
DEBUG hyper::proto::h1::io: flushed 121 bytes

Wrapping up

And that's a wrap! We've got Rust running out in the wild in the form of a serverless function and I couldn't be happier. There's quite a bit of cleanup we could do, for instance handling cases where the repository doesn't exist, but I'll leave that as an exercise for the reader.

All the source code for this function can be on my GitHub here.

Until next, friends!

Going serverless with Rust and Shuttle

What the deployment story currently looks like

Getting started

Managing secrets

errors.rs

handlers.rs

Deploying to shuttle

Back to business

Propagating errors

errors.rs

So many stars

Wrapping up

Not currently listening